International Journal of Computer Applications (0975 - 8887) Volume 58 - No. 15, November 2012 Face Detection and Tracking in Video Sequence using Fuzzy Geometric Face Model and Motion Estimation P. S. Hiremath Department of Computer Science Gulbarga University, Gulbarga-585106 Karnataka, India Manjunath Hiremath Department of Computer Science Gulbarga University, Gulbarga-585106 Karnataka, India Mahesh R. Department of Computer Science Gulbarga University, Gulbarga-585106 Karnataka, India maheshswamy99@gmail.com ABSTRACT With advances in computing and telecommunications technologies, digital images and video are playing key roles in the present infor- mation era. Human face is an important biometric object in image and video databases of surveillance systems. Detecting and locating human faces and facial features in an image or image sequence are important tasks in dynamic environments, such as videos, where noise conditions, illuminations, locations of subjects and pose can vary significantly from frame to frame. In this paper, a novel ap- proach of the detection and tracking of face in video sequence based on the fuzzy geometrical face model and motion estima- tion is presented. The feature extraction process is performed in the support region which is determined by the fuzzy rules to detect face in an image frame. Then, the consecutive frames from a video and their corresponding optical flow are estimated, which are used for tracking face in the video sequence. The experimental results demonstrate the efficacy of the proposed method. Keywords: Face detection, Fuzzy geometric face model, Motion estimation, Tracking.ifx 1. INTRODUCTION With advances in computing and telecommunications tech- nologies, digital images and video are playing key roles in the present information era. The huge amount of visual information is handled by image and video databases, which require effective and efficient mechanisms to index and search these imagery data. In recent years, techniques have been proposed allowing users to search images by visual features, such as texture, color, shape, and sketch, besides traditional textual keywords. Human face is an important biometric object to be searched in image and video databases of surveillance systems. Since face is a unique feature of human beings, and is ubiquitous in photos, news video, documentaries, etc., faces can be used to index and search the image/video databases, classify video scenes, and segment human objects from the background. Therefore, research on face detection has far reaching consequences in image and video database applications. In general image and video databases, however, there is little or no constraint on the number, location, size, and orientation of human faces in the scenes. The backgrounds of these images and video sequences are usually cluttered. Thus, successful face detection and tracking becomes important and challenging before the indexing, search, and recognition of the faces could be done. Detecting and locating human faces and facial features in an image or image sequences are important tasks in dynamic environments, such as videos, where noise conditions, illumina- tions, locations of subjects and pose can vary significantly from frame to frame. A survey of literature on the research work focusing on various potential problems and challenges in the face detection and tracking can be found in [1,2].There have been various ap- proaches proposed for face detection, which could be generally classified into four categories[3]: template matching based methods, feature-based methods; knowledge-based methods, and learning based methods. Template matching based method means the final decision comes from the similarity measurement between input image and the template. It is scale-dependent, rotation-dependent and computationally expensive. Feature- based methods use low-level features such as intensity[4], color[5], edge, shape[6], and texture to locate facial features, and further, find out the face locations. Knowledge-based meth- ods[7] detected an isosceles triangle (for frontal view) or a right triangle (for side view). Learning based methods use a lot of training samples to make the classifier to be capable of judging face from non-face. Despite of the notable successes achieved in the past decades, making a tradeoff between computational complexity and detection efficiency is the main challenge. Among the face detection algorithms, skin color based detec- tion information is an important category[8]. Hiremath and Manjunath[9] have proposed fuzzy geometric approach for face model construction based on only two features, namely, eyes and mouth, for still images which are shown to be the optimal discriminating features for face detection. In this paper, a novel method is proposed for face detection and tracking in a video sequence by using fuzzy geometric face model and motion estimation. The effectiveness of the proposed method is demonstrated by the experimental results. The experi- mentation has been done using publicly available video database. 2. MATERIALS AND METHODS The Honda/UCSD Video Database provides a standard video database for evaluating face detection, tracking and recognition algorithms. Each video sequence is recorded in an indoor environment at 15 frames per second, and each sequence lasted for at least 15 seconds. The resolution of each video sequence is 640x480. Every individual is recorded in at least two video sequences. In each video, the person rotates and turns his/her head in his/her own preferred order and speed, and typically in about 15 seconds, the individual is able to provide a wide range of different poses. The Honda/UCSD Video Database contains two datasets. The first dataset is recorded by a SONY EVI-D30 camera at Honda Research Institute in 2002. It includes three different subsets, one each for training, testing, and occlusion testing. Each subset 12